Setting priorities to address the research gaps between
agricultural systems analysis and food security outcomes in low- and middle-income
countries
Working Paper No. 255
CGIAR Research Program on Climate Change, Agriculture and Food Security (CCAFS)
Charles F. Nicholson
Emma C. Stephens
Andrew D. Jones
Birgit Kopainsky
David Parsons
James Garrett
Setting priorities to address the research gaps between agricultural systems analysis and food security outcomes in low- and middle-income countries
Working Paper No. 255
CGIAR Research Program on Climate Change, Agriculture and Food Security (CCAFS)
Charles F. Nicholson
Emma C. Stephens
Andrew D. Jones
Birgit Kopainsky
David Parsons
James Garrett
Correct citation:
Nicholson CF, Stephens EC, Jones AD, Kopainsky B, Parsons D, Garrett J. 2019. Setting priorities to address the research gaps between agricultural systems analysis and food security outcomes in low- and middle-income countries. CCAFS Working Paper no. 255. Wageningen, the Netherlands: CGIAR Research Program on Climate Change, Agriculture and Food Security (CCAFS). Available online at:
www.ccafs.cgiar.org
Titles in this Working Paper series aim to disseminate interim climate change, agriculture and food security research and practices and stimulate feedback from the scientific community.
The CGIAR Research Program on Climate Change, Agriculture and Food Security (CCAFS) is a strategic partnership of CGIAR and Future Earth, led by the International Center for Tropical Agriculture (CIAT). The Program is carried out with funding by CGIAR Fund Donors, Australia (ACIAR), Ireland (Irish Aid), Netherlands (Ministry of Foreign Affairs), New Zealand Ministry of Foreign Affairs & Trade; Switzerland (SDC); Thailand; The UK Government (UK Aid); USA (USAID); The European Union (EU); and with technical support from The International Fund for Agricultural Development (IFAD). For more information, please visit https://ccafs.cgiar.org/donors.
Contact:
CCAFS Program Management Unit, Wageningen University & Research, Lumen building, Droevendaalsesteeg 3a, 6708 PB Wageningen, the Netherlands. Email: [email protected]
Creative Commons License
This Working Paper is licensed under a Creative Commons Attribution – NonCommercial–NoDerivs 3.0 Unported License.
Articles appearing in this publication may be freely quoted and reproduced provided the source is acknowledged. No use of this publication may be made for resale or other commercial purposes.
© 2019 CGIAR Research Program on Climate Change, Agriculture and Food Security (CCAFS).
CCAFS Working Paper no. 255
DISCLAIMER:
This Working Paper has been prepared as an output for the CCAFS Priorities and Policies for CSA Flagship under the CCAFS program and has not been peer reviewed. Any opinions stated herein are those of the author(s) and do not necessarily reflect the policies or opinions of CCAFS, donor agencies, or partners.
All images remain the sole property of their source and may not be used for any purpose without written permission of the source.
Abstract
This document assesses the current state of practice for the representation of food security indicators in agricultural systems models and provides recommendations for improvements in both model formulation and the empirical evidence base underlying it.
This assessment was based on a review of existing conceptual frameworks linking agriculture and food security, the indicators most commonly used to represent food security dimensions (availability, access, utilization and stability) and studies using models to assess household and regional food security. We also undertook proof-of- concept analyses using household-level and regional-level models incorporating food access indicators into two agricultural systems models.
We found that there is a broad agreement at a conceptual level about important linkages between agricultural systems and food security, at least for some populations. Despite this consensus, the extant conceptual frameworks often are not specific enough about both food security indicators and linking pathways to provide guidance for the integration of food security into agricultural systems models. Our review of the Food Environments literature indicates that it currently emphasizes a broad range of
environmental and personal factors that influence food choice in higher-income country settings, but additional work is necessary to apply these concepts to low- and middle- income countries, and to populations of agricultural producers.
The representations of food security indicators in empirical model analyses of both households and regions are diverse yet often inconsistent with the definitions more commonly emphasized by human nutritionists. Often, empirical models appear to equate measures of production or yields with “food security” when these are indicators only of the “availability” dimension of food security. In general, agricultural system model analyses more commonly employ availability indicators (which can be viewed as a necessary but not sufficient condition for “food security”) but would provide improved guidance for research and programmatic efforts with a focus on indicators of food access. Even when dynamic models are specified, the time units, time horizons and criteria to evaluate the “stability” dimension of food security often are not adequate.
We recommend that agricultural systems models focus on incorporating three food access indicators: 1) food consumption expenditures, 2) experience-based food
insecurity scales such as the Food Insecurity Experience Scales (FIES) or the Household Food Insecurity Access Scale (HFIAS) instruments, and 3) measures of household dietary diversity such as the Household Dietary Diversity Score (HDDS). These indicators are preferable because of the limited empirical relationship between national-level availability and individual nutritional status and because capturing own production on farms or production at regional scales is not sufficient for understanding households’ and individuals’ experience of food insecurity, which entails considerable access to markets, dependence on food prices, and interactions with diverse food environments. Moreover, these indicators should also be evaluated over time using the approaches like that developed by Herrera (2017) to assess more formally the
robustness and adaptability components defining food security stability.
The evidence base is currently insufficient to support robust and reliable integration of experience-based food insecurity scales and household dietary diversity into
agricultural systems models. Although a number of studies have examined the
determinants of these indicators and found a few consistent relationships (e.g., higher household incomes improve all indicators) often these are not specific to the settings modeled by existing agricultural systems models. This suggests that collection of this information, preferably using longitudinal data approaches, is needed so that model extensions can include these indicators.
Additional study (implying larger and longer-term investments) is needed to document and refine the general nature of relationships between common outputs of agricultural systems models and the other two indicators of food access (food insecurity and
household dietary diversity scales). There is also undoubtedly much work to be done to determine appropriate analytical (statistical) techniques, theoretical foundations and functional forms linking determinants to these and other indicators for the purposes of agricultural systems modeling but even more simplistic, reduced form empirical relationships may be useful as this body of work is explored and expanded.
Priorities for application of agricultural systems models integrating improved representations of food security indicators could include assessment of shocks that
could negatively affect production or incomes (e.g., weather, pests, disease, rapid changes in market conditions or access). Other key assessments could include longer- term processes that could negatively affect food security such as climate change (both effects of changes in rainfall and temperature distribution and evaluation of adaptation strategies), land use change, land fragmentation (or consolidation policies), decreases in biodiversity, natural resource degradation and demographic shifts (migration to urban areas).
Our proof-of-concept analyses incorporating food access indicators at the household and regional levels have highlighted the empirical challenges of doing so, but also the
benefits of doing so. For example, the household-level analysis using the CLASSES model indicated that for two different households, food security outcomes are not
“robust” with respect to a yield shock but demonstrate “adaptability” in returning to close to pre-shock conditions. The CLASSES model also indicates the desirability of incorporating multiple alternative measures of food security, because these respond differently over time in the face of a shock.
We recommend broad dissemination of the findings of this study to the agricultural systems modeling community and to the nutrition community (those working in the agriculture-nutrition space in particular). We encourage investments to support development of a broader base of empirical evidence about the determinants of food access indicators and their linkages to variables included in agricultural systems models, and efforts to extend existing agricultural systems models to include improved
representations of food access indicators and intra-household food security outcomes.
Moreover, further assessment is merited of the costs and benefits of representing utilization indicators (such as nutritional status) in agricultural systems models.
Keywords
Farming systems; Food access; Food security; Mathematical models.
About the authors
Charles F. Nicholson is an Adjunct Associate Professor in the Charles H. Dyson School of Applied Economics and Management and International Professor of Applied Economics and Management, International Programs, College of Agriculture and Life Sciences, Cornell University. Contact: [email protected]
Emma C. Stephens is a Professor of Economics at Pitzer College. Contact:
Andrew D. Jones is the John G. Searle Assistant Professor of Nutritional Sciences, School of Public Health, University of Michigan, University of Michigan. Contact:
Birgit Kopainsky is a Professor in System Dynamics at the Department of Geography, Bergen University. Contact: [email protected]
David Parsons is a Professor at the Swedish University of Agricultural Sciences, Department of Agricultural Research for Northern Sweden; Crop Production Unit.
Contact: [email protected]
James Garrett is a Senior Research Fellow at Bioversity International. Contact:
Acknowledgements
The authors thank Phillip Thornton and Laura Cramer of CCAFS for their support during the development and implementation of this project. We also greatly appreciate the extensive and insightful comments on the first draft of this working paper provided by Jessica Bogard, Nutrition Systems Scientist with CSIRO Agriculture and Food. We also gratefully acknowledge the support of the System Dynamics Group in the Department of Geography at the University of Bergen for hosting a meeting of the authors in June 2018.
Contents
1. Introduction 9
2. Scope of the assessment 11
3. Review of existing conceptual frameworks linking agricultural systems and food
security 14
4. Review of existing quantitative systems models linking agricultural systems and
food security 28
5. Review of food security indicators 44
6. Priority research themes and settings for integration of food security indicators into
agricultural systems models 54
7. Proof-of-concept case analysis for integration of food security indicators into
agricultural systems models 71
8. Overall recommendations: the way forward for improved integration of food
security indicators into agricultural systems models 98
References 103
1. Introduction
The linkages between agriculture and nutrition-related outcomes—including food security outcomes—have long been recognized in various conceptual frameworks.
Actions based on these linkages have become more prominent during the past decade with efforts such as the United Nations Scaling Up Nutrition and other organizational efforts to “mainstream nutrition” into sectors beyond health (IFAD, 2014). In particular, nutritional considerations have become more important in the design and
implementation of agricultural development projects and best practices have been proposed (e.g., FAO, 2013; Garrett, 2017). Although agriculture is only one among many factors influencing food security outcomes, for certain populations and regions the linkages between food security outcomes and the performance of agricultural systems are vitally important—particularly in light of ongoing environmental challenges due to soil degradation, water availability and global climate change.
Despite the recognition of these important linkages and challenges, there are a limited number of studies that include explicit quantitative analysis of the linkages between food security outcomes and agricultural systems. In a review of previous research for a special issue of Agricultural Systems of papers compiled from the 2nd International Conference on Global Food Security in 2015, Stephens et al. (2018) noted the gap between conceptualization and quantitative implementation of linkages between agricultural systems outcomes and food security, stating:
An emphasis on measuring household or individual level access to food, and
understanding the dietary or nutritional impacts of changes to agricultural systems are conspicuously underrepresented…
They ultimately concluded that:
…further work is needed to examine the interfaces between agricultural systems, food systems and food security, including examination of agricultural produce markets, value chains, international exports and imports of agricultural commodities, food demand and preferences and constraints (so called ‘food environments’ Herforth and Ahmed, 2015).
A few studies (e.g., Stephens et al. 2012, Kopainsky and Nicholson, 2015) have tried to link agricultural systems models with food security outcomes to understand evolving intertemporal dynamics and assess the impacts of system intensification. However, such studies appear to be small in number and are limited by the use of rudimentary indicators of food security (e.g., proportion of household caloric needs) and a focus only on household-level outcomes.
Thus, there is a crucial need for—and a large potential benefit to—additional exploration of the “uncharted territory” (Stephens et al., 2018) linking agricultural systems analysis and food security outcomes in a broader sense. This paper provides a further update on the current state of literature encompassing quantitative linkages between agricultural systems analysis and food system outcomes, identifies priority research actions for improving the quantitative analysis of such linkages at household and regional scales, and illustrates how the integration of food security indicators into agricultural systems models might be done with a proof-of-concept case analysis.
Objectives
This working paper has the following objectives:
1. Additional review and assessment of systems-oriented conceptual frameworks that link food security outcomes to other components of agricultural systems, building on the discussion in Stephens et al (2018);
2. Additional review and assessment of previously-developed quantitative models that link agricultural system outcomes and food security outcomes, also building on the discussion in Stephens et al (2018);
3. Delineation of priority research themes and contexts that would facilitate analysis of key linkages between quantitative agricultural systems analysis (with an emphasis on systems modeling) and a relevant set of food security outcomes at household and regional scales;
4. Describe a proof-of-concept case analysis illustrating the process of integration and the usefulness of explicit consideration of linkages.
2. Scope of the assessment
A few clarifications and caveats are appropriate to more clearly delineate the scope of this assessment. A principal purpose of this document is to provide guidelines and recommendations for improvement of the practice of modeling food security outcomes using agricultural systems models. Although not categorically excluding other types of analyses from our discussions, we generally imply by the term ‘agricultural systems model’ an empirical model that includes biophysical content, sometimes complemented by economic content because both of these elements can be necessary (if not sufficient) for an assessment of linkages between agriculture and food security indicators. Often, this will comprise a simulation model (of one or more types) that is used for the
assessment of counterfactual situations compared to a baseline or status quo situation—
in contrast to a purely statistical model that is used primarily to determine the nature of associations between variables1. Although our assessment of the literature has turned up many types of models, our focus in this assessment is on how to better represent food security outcomes in those models fitting our definition of an ‘agricultural systems model’. The extensive literature on agriculture, food security, and systems models required us to impose some limits on our review.
Many agricultural systems models focus at the plot, farm, household or landscape level due to their focus on biophysical dimensions of agricultural production. In contrast, many models assessing food security outcomes tend to be focused on the household or on national or regional markets. In general, our focus is on food security outcomes for households that have an active role in agriculture, rather than for all households in a given region. This is consistent with common practice for household-level agricultural systems models, as illustrated by analyses such as Stephens et al. (2012) and Wossen et al. (2018). However, there are examples of analyses that integrate households across regional markets (e.g., the agent-based modeling work of Bakker et al., 2018) that could
1 We acknowledge that some studies (i.e., Harttgen et al., 2016) develop simulations based on a previously- estimated statistical model, but most simulation models use a variety of relationships that are not purely
readily be extended to assess the impacts on non-agricultural households (e.g., urban residents not producing their own food). Despite our focus on agricultural households so that biophysical, economic and food security outcomes can be more closely linked, the basic approach we employ could in principle be applied to other simulation model settings at various scales, including analysis of non-farming households for whom representing consumption decisions would be most relevant. We define “regional” as a higher level of aggregation than an individual household, which can encompass various spatial aggregations (typically, at the level of a country or its subregions).
Our focus on agricultural systems models and food security has a number of
implications. First, although there is broad and continuously-growing literature on the linkages between agriculture and food security or food and nutrition security (reflected, e.g., in the development of ‘nutrition-sensitive agriculture’ and related analyses), we generally limit our review to those analyses that have been formalized in development of empirical (and simulation) models. The broader literature linking agriculture to food security outcomes can be a valuable complement to the development of improved agricultural systems models, but we deemed a comprehensive review of this literature as outside of the scope of this document.
We have provided only cursory treatment of linkages between agricultural systems and intra-household (individual) food security outcomes, despite its acknowledged
importance, particularly for women and children. We have done so in part because of the quite limited treatment of intra-household outcomes in the existing agricultural systems modeling literature, and because we believe additional assessment of the costs and benefits of alternative approaches to modeling intra-household disaggregation is merited. We offer some assessments of the current state of practice of intra-household representations throughout.
Finally, although food security frequently is defined to include four elements
(availability, access, utilization and stability) we focus much of our discussion on the access and stability dimensions. As we note below, the availability component is often the most easily measured and represented in agricultural systems models, but improved availability should generally be thought of as necessary but not sufficient for improved
food security, given the somewhat hierarchical nature of these four elements2. Thus, we believe it is both necessary and useful for agricultural systems models to transcend the use of only availability measures. Utilization typically comprises food actually
consumed by individuals and the resulting individual nutritional outcomes. Often, these outcomes are described as related to “food and nutrition security” (FNS), which
certainly has considerable overlap with our treatment of “food security.” However, because the utilization component often has substantive interactions with health status (see, e.g., Randolph et al., 2007) that are challenging to represent in agricultural systems models, we do not focus on the utilization component of food security. However, we note in the conclusions some recommendations for follow-on work that could encompass the broader concept of FNS.
2 For example, access will necessarily be restricted without adequate availability, but increased availability (say, through increased production) does not imply that access will be improved for a substantive number
3. Review of existing conceptual frameworks linking agricultural systems and food security
Conceptual frameworks that link food security outcomes to other components of agricultural systems provide a starting point for examining research gaps between agricultural systems analysis and food security outcomes. There is a large and growing literature that hypothesizes and documents the linkages between agriculture, nutrition and health. Our objective here is to review the conceptual bases that have been offered for these linkages, with two main purposes related to our assessment of food security outcomes using agricultural systems models. First, a review of the conceptual basis indicates the degree of consensus regarding the underlying nature of these
relationships, which can be used to motivate their explicit inclusion in agricultural systems models. Second, this review provides a means of reviewing hypothesized pathways and effects that may be useful to guide the development of agricultural systems models with explicit linkages to food security for specific contexts. Thus, we undertook a review of a variety of conceptual frameworks, including those that focus on food and nutrition security as well as those that represent a more general “food system”.
We began reviewing frameworks with which the authors were already familiar and additional relevant frameworks were identified in SCOPUS using the search terms “food security conceptual framework.” We also offer comments on how the existing
frameworks might be modified or complemented to facilitate their use in the development of quantitative (especially structural) modeling approaches.
The literature on conceptual frameworks that link agriculture with food security is growing, and early frameworks that differentiate between food system activities, outcomes and drivers (cf. Ingram et al., 2010) are being refined (e.g., by the explicit discussion of the role of diets as a core link between food systems and their nutrition and health outcomes (HLPE, 2017) and extended (e.g., by the explicit discussion of the political system and governance issues, e.g., Braun & Birner, 2017; Wegener et al., 2012).
Existing conceptual frameworks that link food security outcomes to other components of agricultural systems share a number of features and components. Many frameworks acknowledge that food systems are complex and adaptive systems that are composed of:
▪ Food system activities such as food production, processing, distribution, and consumption.
▪ The resources going into these activities.
▪ Outcomes of these activities, spanning from food security to environmental and social welfare outcomes.
▪ Actors, institutions, and organisms whose decentralized behavior and interaction shape and modify food system activities and resource use and whose behavior and interaction might change in response to food system outcomes.
▪ Feedback and interdependence across levels and scales.
Nearly all of the frameworks recognize that a wide variety of factors—not just
agriculture—affect food security outcomes for both households involved in agriculture and those that are not. More recent additions to existing frameworks are the concepts of food environments and resilience. Food environments describe the physical, economic, political and socio-cultural context in which consumers engage with the food system to acquire, prepare and consume food (HLPE, 2017). Resilience refers to the capacity of the food system to provide food security over time and despite disturbances (Tendall et al., 2015). There are three generic potential responses for food systems when they are affected by disturbance (Walker et al., 2004):
▪ Stability or robustness: the system does not exhibit changes in its behavior. Stability describes a behavior that follows the same trajectory as it would without a
disturbance.
▪ Adaptation: the behavior of the system bends when affected by a disturbance but eventually, it bounces back to the behavior over time of a system without a disturbance.
▪ Transformation: the system as it currently exists breaks and changes into a new system with different structure, relationships and identity. The new system might or might not produce the same outcomes (e.g., food security). Whereas some transformations might be positive, risk management is often concerned with those transformations that are not and with cases in which the system might collapse.
In general, the above criteria as defined are most relevant for the consideration of (often, unintended) shocks that would have a negative impact on food security (such as drought, pests, disease or conflict). For the analysis of (often, intended) interventions to improve food security outcomes (such as productivity-enhancing technologies),
“stability” (no change) or “adaptation” (return to previous conditions) would generally be considered less than desirable. (We illustrate this with our proof-of-concept analysis with the Mexico Sheep Sector Model, p. 65 below.) This suggests the need to more carefully define the meanings of these indicators with respect to the analyses to be undertaken, in addition to more clearly defining what constitutes “stability” and
“adaptation.” In addition, it is generally more challenging to assess the “transformation”
component in agricultural systems models, and this appears less common in the literature we review below. Although dynamic models (perhaps most particularly agent-based model analyses) could in principle capture some types of transformative change, it may be adequate for analyses with a time frame extending to only a few years to focus on the first two of these responses to system shocks or evolution.
Herrera (2017) develops a series of metrices that can be calculated with dynamic simulation models to assess stability or robustness, adaptation and transformation in social-ecological systems. The metrices help a) anticipating whether robustness, adaptation or transformation can be expected as a result of a given disturbance, b) identifying where the thresholds are between robustness, adaptation and
transformation and c) understanding what the resources and drivers are that foster robustness, adaptation and transformation. The metrices described in Herrera (2017) all refer to the impact of a disturbance (defined as the multiplication of the extent of a shock and the duration thereof) with respect to an outcome function. The outcome function describes the behavior over time of variables or indicators of interest such as food security indicators. The impact of a disturbance is usually measured by comparing the time-dependent behavior of the outcome function with the reference behavior of the same function, that is, with the time-dependent behavior of the outcome function in the absence of a disturbance. Four main resilience metrices discussed in Herrera (2017) are:
▪ Hardness: The ability of the system to withstand a disturbance without experiencing a change in the performance of the outcome function F(x) (the threshold value between robustness and adaptation).
▪ Recovery rapidity: The average rate at which the system returns to the reference behavior of the outcome function (i.e., returns to the same steady state, pathway or regime).
▪ Elasticity: The ability of the system to recover from a disturbance without changing to a different steady state or regime (the threshold value between adaptation and transformation).
▪ Index of resilience: The probability of keeping the current steady state or regime.
Hardness and elasticity indicators are examined more specifically for the two proof-of- concept models (in Section 7).
Appendix 1 provides a more detailed overview of the conceptual frameworks in diagrams and tabular form (Appendix Table A1). The myriad of frameworks seems to serve different purposes:
Some provide a high-level perspective on the interconnected nature of agricultural systems and food-related outcomes. These frameworks illustrate that food security both depends on and influences agricultural systems. Examples of this type of conceptual frameworks are: Fanzo et al., 2017; IOM, 2015; Neff & Lawrence, 2015; Sobal et al., 1998.
Another set of frameworks provides more details about the connections between environmental, farming, economic and social sectors. They identify and visualize the major subsystems and key connections among them. Examples of this type of conceptual framework are in Fanzo, et al., 2017; HLPE, 2017; Horton et al., 2016;
Ingram, et al., 2010; Pinstrup-Andersen & Watson II, 2011; Wegener, et al., 2012.
A last category of frameworks has a somewhat narrower focus but describes the specific pathways linking agricultural systems and food and nutrition security. Examples of this type of conceptual frameworks are: Acharya et al., 2014; Hammond & Dubé, 2012; K.
Suneetha et al., 2014; Kanter et al., 2015.
Maybe the most comprehensive effort at conceptualizing the linkages between
agricultural systems and food and nutrition security is the global food system map that depicts the inter-related concepts and challenges that connect the global food system (Figure 1; ShiftN, 2009).
Discussion of existing frameworks to support modeling linkages between agricultural systems and food security
Many of the frameworks discussed above provide insights about how to model the linkages between agricultural systems and food security. The most useful for the purposes of systems model development tend to be those that focus on food security and specify pathways linking agriculture to food and nutrition outcomes. These include frameworks presented in Kadiyala et al (2014), Randolph et al. (2007), Dobbie and Balbi (2017), Garrett (2017), Kanter et al. (2015) and Sassi (2018). The illustrative pathways in these frameworks suggest more directly the mechanisms (variables and
relationships) by which agricultural systems outcomes and food security outcomes are linked. Many of the frameworks are quite high-level and describe very general
relationships rather than specific pathways. Perhaps the most notable example is from Wossen et al (2018), for which “Adaptation” is directly linked to “Food Security” in one- way causality. These higher-level depictions can be useful as conceptual guidelines, but they provide limited support for quantitative model development and assessment of interventions because they are not sufficiently specific about quantitative indicators and impact pathways. (In some cases, “policy” is viewed as a higher-level determinant of food security, but simply stating that is not sufficiently specific to provide insight about how to change policy.) The ShiftN (2009) food system diagrams have a greater level of complexity and begin to delineate pathways, but they don’t really focus clearly on food security.
Most of the frameworks (even some that focus on food security) do not include all elements of availability, access, utilization and stability. Especially the latter is more frequently ignored, as discussed further below. In addition, it is often not clear if these are viewed as some sort of hierarchy (especially the availability-access-utilization linkages) or whether they are separate. In some cases, access causes availability in a
diagram, in other cases, it is the reverse. Related to this is the frequent absence of delineating levels of analysis (, data or outcomes. Most of the frameworks also do not include specific indicators for food security or nutrition outcomes. It is common to have the outcome be “food security” or “nutritional status” and only a few mention specific indicators at the household level such as dietary diversity (e.g., Kanter et al, 2015). This higher-level approach may be appropriate for the intended purposes of the frameworks, but they may not provide much guidance to quantitative model developers.
Many frameworks are also not particularly clear about which actors are covered and who makes what decisions. This is relevant models often need to specify one or more decision makers at multiple scales. The Hawkes (2009) and Hawkes et al. (2012) frameworks use an Actors-Processes-Outcomes framework, but this is quite high level and “processes” include “ag inputs” that are not always clearly defined. Arachya et al.
(2014) includes producers, “food chain actors” and consumers. “Consumers” or
“households” are frequently represented (e.g., Garrett, 2017; Ecker and Breisinger, 2012). Sometimes the frameworks delineate “levels” (e.g., national, regional,
community, household, individual) with specific effects or outcomes of interest for each (e.g., the Food Insecurity and Vulnerability Information and Mapping System (FIVIMS), FAO, 2000).
Few of the frameworks address intra-household food security issues. Of the more than 50 frameworks reviewed (and summarized in Appendix 1), only 4 had explicit
treatment of individuals with the household, focused on children (especially for nutritional status) and women. Three other frameworks implied treatment of
individuals (e.g., Sassi 2018 mentions “individual food and nutrition pathways”) but in general the conceptual treatment of the linkages determining intra-household food security status is limited. Although we acknowledge that we did not search for frameworks specifically addressing intra-household allocation and outcomes, the limited treatment of this issue in more general frameworks suggests the need for a reconsideration of the treatment of intra-household issues from both the conceptual and empirical perspective.
Most of the frameworks do not specifically represent intertemporal dynamics or feedback processes, both of which would be important to represent the “stability”
component of food security. “Stability” is also at times referred to by the broader concept of “resilience”. Intertemporal change is admittedly a challenging concept to represent in a two-dimensional diagram, but improvements to existing frameworks would seem possible in this regard. Some frameworks discuss general resilience concepts (e.g., Tendall et al., 2015; FAO, 2016), but the linkages to the stability component of food security are not explicit. Burchi et al. (2011) depict stability in a framework that primarily defines the four components of food security but include suggested actions and strategies to promote stability of food availability, access and utilization. Allen and Prosperi (2016) integrate resilience concepts into the Ericksen (2008) and Ingram (2011) frameworks.
Many of the frameworks also depict a linear cause-and-effect model with limited
feedbacks among system elements determining food security outcomes. Representation of feedback is relevant because—as noted above—these systems demonstrate feedback and interdependence within and across levels. Appropriate representation of feedback processes is relevant, particularly when considering proposed agriculture-based interventions designed to improve food security outcomes. The systems modeling literature (e.g., as summarized in Sterman, 2000) has long since noted that feedback processes, accumulation and non-linearities result in “dynamic complexity”, which gives rise to “policy resistance” (the intended effects of interventions will be delayed or largely offset) and “unintended consequences” (other, often negative, effects may occur in response to interventions; short-term and long-term impacts of system changes can differ). Thus, understanding and appropriately representing feedback processes in conceptual frameworks and quantitative models will often be both necessary and appropriate. Moreover, feedback representations provide a specific link with
intertemporal dynamics that is often appropriate, as noted above. Most intertemporal quantitative models include at least some feedback processes that link system elements over time, so an understanding of which feedback processes are likely to be important conceptually is relevant for empirical model development (including data collection efforts).
The frameworks that do represent feedback processes tend to include only a few such linkages that differ for each diagram. General resilience frameworks (e.g., IOM 2015;
FAO, 2016, Tendall et al. 2015) tend to represent changes in high-level “state” (key variables) over time. The high-level framework from Hammond and Dube (2012) indicates feedback processes (and some specific mechanisms) among the “agri-food”,
“environmental” and “health/disease” components of the system that determines food and nutrition security. One of the more common inclusions is feedbacks between the food system (or agriculture) and environmental outcomes (Lawrence, 2015; Horton et al., 2016; Burchi et al., 2011; Ericksen, 2008; Ingram 2011; Allen and Prosperi, 2016;
ShiftN, 2009). Frameworks that focus on household assets and livelihood strategies (e.g., Kadiyala et al., 2014; Ashley and Carney, 1999; World Food Programme, 2012) tend to link livelihood outcomes (including food security) back to increases in household assets in a reinforcing feedback loop. Similarly, the UNICEF (1998) framework shows a
reinforcing feedback process where lack of initial livelihood assets limits improvements in child nutritional status—with ongoing intertemporal effects.
Other frameworks focus on feedbacks between consumer decisions and the structure of food supply chains and food environments (e.g., Pinstrup-Andersen and Watson, 2011;
HLPE, 2017; Arachya et al, 2014; Hawkes et al. (2012). An extension of this concept includes when consumer decisions and related outcomes (nutritional, social, economic, environmental) are hypothesized to affect system drivers such as biophysical,
environmental, technology, political, socio-cultural, and demographic ones (as in HLPE, 2017; Ericksen, 2008; Ingram, 2011; Allen and Prosperi, 2016). More specific to food security, a number of frameworks depict interactions—if not exactly feedback—
between nutrition outcomes and health outcomes (Garrett, 2017; et al., 2012; Randolph et al., 2007).
Although all of the represented feedback processes are likely to be appropriate for specific purposes, the lack of consistency among the frameworks implies challenges for effective representation of these effects in agricultural systems models linking to food security outcomes. The Randolph et al. (2007) diagram is probably the most detailed and relevant of the feedback-inclusive frameworks, because it provides a more detailed representation of alternative pathways (including some described elsewhere, e.g., Kadiyala et al., 2014; Gillespie et al, 2012) linking agriculture, nutrition and health for the specific context of livestock ownership.
Not surprisingly, diagramming conventions are highly variable. Many frameworks show connecting lines (sometimes with arrows in both directions) without really indicating implied directions of causality, and only Randolph et al (2007) indicates polarities of hypothesized linkages. Diagrams are inconsistent in their depictions of hypothesized feedback processes, and in some cases it is difficult to determine what is connected to what. Language is often cryptic or a bit inconsistent among linked variables (e.g.,
“resources” cause “inadequate education”). The conventions used in “Causal Loop Diagramming” (e.g., Sterman, 2000) and similar hybrid diagrams that also show stocks and flows would bring a good deal of additional clarity of meaning to these diagrams (and allow them to more clearly delineate hypothesized pathways).
Many of the frameworks could also more clearly delineate so-called “model boundaries”, which define what is endogenous and what is exogenous for the purposes of the
(conceptual or quantitative) analysis. In many frameworks, “context” or “environment”
variables appear to be assumed to be exogenous, and these encompass a vast variety of factors (political, social, cultural, knowledge, infrastructure, services, (macro)economic, climate, disease outbreak, policies, programs, conflicts, technology, food environments, legal systems, ethical values, productive assets and sometimes even food availability itself). For the purposes of many of frameworks, assuming these to be exogenous may be fine, but from a modeling perspective delineation of the model boundary is
important. It is also not clear at what level many of these factors have the largest influence on outcomes. For example, the WFP framework suggests that all factors have equal impact at the community and household levels, and ‘exposure to shocks and hazards’ affects all levels (implied equally). This also doesn’t suggest much to modelers about which effects or causal relationships are most important.
Figure 1: Global Food System Map
3. Source: ShiftN, 2009
Discussion of food environments literature in the context of
agricultural systems and food security modeling in low- and middle- income countries
A growing number of studies have more recently applied the ‘ecological system theory approach’ (e.g., Bronfenbrenner, 1989) from the human development literature to the analysis of food system outcomes (e.g., Herforth and Ahmed, 2015), which highlights the increased understanding of the importance of the food environment. . This subfield conceptualizes food acquisition and consumption choices and opportunities as being driven and shaped by what has been classified as an individual’s ‘food environment.’
This food environment is often defined—in conceptual terms at least—rather broadly as all factors affecting choices about the consumption of food. This includes factors such as the spatial density of foods on offer, food prices, product properties (e.g., quality, safety, convenience, diversity), the types of vendors offering food and “food messaging” such as advertising and promotion (HLPE, 2017). The food environment is frequently conceived
of a set of overlapping hierarchical influences comprising social and cultural norms and values, sectors of influence (e.g., government and media), environmental settings where food is consumed, and individual factors (e.g., demographics and knowledge) that affect food intake and physical activity levels (Herforth and Ahmed, 2015). Given the potential overlap between agricultural systems modeling and frameworks emphasizing the “Food Environment” (FE) as a key determinant of food security outcomes, we examined the current status of the literature on Food Environments to assess its potential relevance.
The diversity of factors characterizing the food environment presents challenges for more complete integration of these concepts into agricultural systems models that would also represent food security outcomes.
To date, much research about food environments has been conducted in high-income regions, typically investigating potential food environment drivers of health issues resulting from over-nutrition, such as obesity. Key metrics in these contexts have included spatial analysis of the location and distance to food sources for certain
populations and communities, the relative affordability of foods with respect to average incomes of consumers, inventories of food types and quality within food source outlets like stores and restaurants, or detailed breakdowns of the nutritional content of foods that are available to a population of interest. Lytle and Sokol (2017) and Ruel et al (2017) recently surveyed the literature and conclude that spatial indicators as food environment metrics dominate (such as the density of food retailers in a city center), partly due to the relative ease of obtaining these data compared to collecting detailed inventories of food outlets. Thus, much of this literature tends to emphasize settings in which food consumption decisions are made by individuals and households that do not produce substantive amounts of the food they consume—which is in contrast to the populations of agricultural producers often represented in agricultural systems models.
A recent brief by the Food Environment Working group on research gaps on food environments emphasizes the need to conduct research to apply the food environment concept in low- and middle-income countries (Turner et al 2017). Work that would emphasize elements of the food environment for households that are food producers (even if net buyers) would be most relevant for linkages to agricultural systems models, because farm production and sales patterns would be a major influence on the types of
food available, distribution, relative prices and food quality. The inclusion of food security metrics into agricultural systems models in recommended in this paper, such as food consumption expenditures or the household food insecurity scale, would go some way to filling this void by linking food choices to food production in low- and middle- income settings.
Given the (incomplete) overlap between typical agricultural systems and food
environment analyses it is useful to assess those variables commonly used in the food environment literature and their potential to be included in agricultural systems models (Table 1). As outlined by Lytle and Sokol (2017), the food environment has thus far been assessed through a host of survey instruments that gather variables that collectively characterize the overall food environment. For example, one common measurement tool is a ‘Market Basket’ questionnaire. With this, researchers estimate the overall price of a common basket of important food goods across multiple food outlets, including unit prices and quality data for a fixed set of items (e.g. milk or dairy, produce, meat etc.). Ranking the cost of a common basket for different target populations is used to assess the food environment, with lower cost baskets serving as a proxy for a better food environment overall. In contrast, agricultural systems models more frequently focus on a few specific commodities. However, market prices and agricultural output quality, of at least the commodities being modeled, are outputs from many agricultural system models that can be used to assess some components of the food environment.
Given the large range of food environment variables that could be considered, this review project will only cover a subset of food environment metrics that we view as more readily able to be incorporated into agricultural systems model analyses.
The Food Environment Working group brief also outlines another useful conceptual framework for us to consider, breaking down the elements of the food environment further into an ‘external’ as well as a ‘personal’ food environment. The ‘external’ food environment often consists of exogenous factors that influence food acquisition and consumption, like spatial indicators of locations of food outlets, but also food prices in markets and food quality properties. The ‘personal’ food environment often consists of endogenous variables that are specific to household food choices, like income and expenditures on food, time constraints to obtaining and preparing food, household
demographic composition and preferences. It seems possible, therefore, that some of these factors may be important, both for our primary objective of jointly modeling agricultural systems and food security indicators (like food prices or distance to markets) but also could help tease out the role of the food environment in the overall food security (and agricultural system) status of a household in a low/middle income setting. Some factors can be treated as exogenous model variables (like prices for subsistence commodities or household size) that could be adjusted to test their
influence over model outcomes. They will be important in determining multiple aspects of the agricultural system behavior, calculating the food security indicator of interest and also can represent important elements of the food environment that vary across low/income settings. However, some other potential indicators, like spatial surveys or checklists of the inventories of types/qualities of foods around a given population (as exemplified by the ‘food desert’4 concept, for example), will be less relevant and would be difficult to validate with data from the low/middle income settings we are
considering (rural mixed farming communities in Kenya, or commercial sheep farmers in Mexico). Thus, an incomplete accounting of the influence or role of the food
environment, as captured by jointly appropriate variables, is the most likely outcome.
4 Food deserts are defined as geographic locations lacking access to fresh fruit, vegetables, and other healthful whole foods, usually found in impoverished areas. This is largely due to a lack of grocery stores, farmers’ markets, and healthy food providers (American Nutrition Association, ANA Nutrition Digest, volume 38, number 2. http://americannutritionassociation.org/newsletter/usda-defines-food-deserts. USDA’s definition is that a “low-access community,” must have at least 500 people and/or at least 33 percent of the census tract's population that reside more than one mile from a supermarket or large grocery store (for rural census tracts, the distance is more than 10 miles).
Table 1. Food Environment Measurement Tools, Variables and Potential for Inclusion in Agricultural Systems Models
Food Environment Measurement Tool, Food Environment Variables Included
Variables with HIGH Potential for Inclusion in Agricultural System Models
Variables with LOWER Potential (or larger difficulty) for Inclusion in Agricultural System Models Interviews/consumer
questionnaires
Eating habits and choices Overall food consumption expenditures
Other aspects of food choices (preferences, tastes, knowledge)
Location of healthy food sources
Comparing on-farm production to market location for non-farm goods via market distance parameters
Healthy food option availability
Assessment of agricultural system output of healthy food vs other commodities (ratios?)
Assessment of overall availability of healthy food in market vs. non-market outlets
Household demographics
Household size and make up (adults vs. kids, gender and food requirements and labor output etc.) as well as food consumption costs
Market basket surveys Unit prices for specific food items and overall basket costs
Market prices (as both drivers of production levels of certain foods, and also as real costs of food baskets overall as a consumption parameter)
Quality ranking for specific items
Quality of food produced – e.g. organic vs. non-organic, nutritional values/contents, ‘improved varieties’
Quality of non-produced items beyond price differentials as a signal of quality
Checklists/inventory analysis
Existence/availability of specific foods in a specific food source
Production choices made for one commodity over another and its importance to food security
Influence of household agricultural systems on total availability of important food commodities Geographic/spatial analysis
Distance between target population and food sources
Inclusion of non-market sources and transactions costs in overall food consumption costs
Sales analysis
Consumer survey of items purchased in a food source vs what is available
Total food consumption expenditures Inclusion of food items not chosen (but available) Nutrient/menu analysis
Consumer survey of items purchased in a food source vs what is available
Assessment of macro/micronutrient content of foods produced/consumed
Nutritional content of available, but not consumed foods in overall environment
4. Review of existing quantitative systems models linking agricultural systems and food security
Basic concepts in agricultural systems modeling
Because non-modelers comprise one of the audiences for this working paper, we provide here a brief introduction to agricultural systems modeling, including a discussion of common general definitions, model types, and concepts related to
household (economic) decision making. We noted above that we generally imply by the term ‘agricultural systems model’ an empirically-based5 model that most commonly includes biophysical relationships (often at the farm or field level) sometimes complemented by economic content. An empirical model specifies mathematically a simplified representation of a specific set of real-world interactions. Often, an
agricultural systems model is a simulation model (of one or more types) that is used for the assessment of counterfactual situations compared to a baseline or status quo
situation—in contrast to a statistical model that is used primarily to determine
associations between observed variables. These models are typically used to predict the impacts of management changes (such as a new crop variety or increased fertilization) or changes in context (e.g., climate or market environment) on outcomes such as crop and livestock yields or production, household incomes and consumption, environmental indicators (e.g., nutrient flows or greenhouse gas emissions) or food availability.
Agricultural systems models are typically represented by a system of equations that describes mathematically the interactions among the different elements of a specific system to be modeled. The model should have a clearly defined model boundary, which indicates the focus of the model’s analytical capability and also what variables are excluded from consideration. It is also important to indicate which variables are
assumed by the model to be endogenous (that is, with values determined by the model’s calculations) or exogenous (with values assumed as inputs, not by the model’s
5 There is a continuum of agricultural systems models that incorporate both empirical and theoretical components. Here we refer to a broad range that have empirical content but exclude those that are primarily or entirely theoretical.
calculations). Exogenous information used in models can include data (such as an assumed time series of prices for an agricultural product) or parameter values (often these are assumed numerical inputs into a calculation in a model equation.)
Agricultural systems models are quite diverse in terms of the agricultural activities and processes they represent (crops, livestock, fisheries, land or landscape management), although it is common for models to focus on a limited number of crop and livestock species—and sometimes their interactions. The scale analyzed can also vary, with models representing the plant, plot, enterprise, farm or household, landscape, region, country or global level. Models also differ in terms of their representation of decisions by a set of actors (often, humans assumed to be managing the system). Some models assume little or no human intervention in the system, whereas others make human decision-making a central component upon which many outcomes depend (see additional discussion below). Agricultural systems models can be static (analyzing a single time period) or dynamic (analyzing multiple time periods, typically with intertemporal linkages among outcomes).
Simulation model is a general term implying use of an empirical model to compare alternative scenarios. A simulation model can focus primarily on biophysical outcomes (such as crop yields or greenhouse gas emissions), economic decision making and outcomes (such as the choice of which crops to plant and determination of household income) or integrate the two kinds of outcomes into a single modeling framework. As an example of this latter type, the CLASSES model (Stephens et al. 2012) has detailed representation of soil nutrients, crop and livestock production, household income and assumes that a household decides how to allocate their resources (land and labor).
Optimization models are typically used to identify what activities will best achieve a desired objective, and often have substantive economic content. An example would be a model to determine the crop mix for given farm would use to generate the largest possible farm income. Optimization models are also used to determine the equilibrium price and quantity outcomes in the markets for one or more crops based on supply and
demand relationships; these are often referred to as partial equilibrium models6. Agent- based models (ABM) include explicit specification of numerous decision-making agents, whose interactions (either through direct sharing of information or through their collective impact on markets) affect outcomes for all of the agents. An example of an ABM is Wossen et al. (2018), who analyzed how multiple households with assumed different characteristics interact through crop and livestock markets to determine incomes and food availability outcomes.
Economic Models and Human Decision Making
Several distinct theoretical approaches and schools of thought about human decision making have emerged from the economics discipline that attempt to explain observed economic decision-making behavior. At present, most researchers in economics employ a variety of mathematical models to represent these theories and capture key aspects of human decision making about scarce resources. In applied settings, researchers use these mathematical representations to explain and analyze empirical data gathered about different economic phenomena, like market trading quantities and prices, or consumer spending patterns, for example.
Two of the main features of economic models of human decision making are:
1) An objective function, which uses a mathematical expression to represent the overall goals and preferences of the decision maker. Examples include utility to describe the overall level of happiness obtained by a consumer after allocating their scarce resources, or profit for a producer in an economic enterprise.
2) A mathematical representation of the constraints (or forms of scarcity) that the decision maker faces, for example, a limited financial budget, available land or labor resources to allocate across different activities.
6 “Partial equilibrium” means that only a limited number of markets (products) are considered in the analysis, whereas “general equilibrium” analyzes explicitly the interactions among all the markets in an economy.
The final piece of economic models of human decision making relates to how economic agents actually make decisions and choices about what to do with their scarce resources.
The predominant theoretical paradigm in the discipline is known as neoclassical economics. Within this, economists assume agents are rational actors, and will make decisions in order to allocate resources in such a way that is optimal from the
perspective of the objective function (as noted above). For example, a consumer is assumed to spend their limited income on consumption goods in order to maximize their utility. This process of optimizing the value of the objective function, while still respecting the constraints, is therefore known as constrained optimization. Thus, for consumers, they make decisions based on constrained utility maximization, producers will make input decisions based on constrained profit maximization (or constrained cost minimization if the objective function for producers is instead to limit their overall costs).
Constrained optimization can be simulated for an economic agent over a variety of important economic parameter values, like prices or income levels. If this is done, the modeler can generate an overall demand function for consumers, that describes what is optimal over a range of circumstances, or a supply function, for a producer, that
indicates a set of profit maximizing choices that the decision maker is assumed to make when economic parameters change.
Within agricultural systems models, approaches to modeling economic factors and decision making vary widely, as initially discussed. In some instances, there is no active decision making done within the model, although the value of an economic objective function, like profits or costs or food consumption, can sometimes be one of the model outputs. The IMPACT model from IFPRI is one large scale example
(https://www.ifpri.org/program/impact-model). In other cases, as in Stephens et al.
(2012), human decision making is actively modeled, with human managers making allocation decisions over scarce resources in order to optimize the value of the relevant objective function (in the CLASSES model, the objective function is economic returns to the farmer’s labor time, which is related to an overall notion of the profitability of labor on the farm).
Important alternatives exist to the neoclassical approach and are sometimes included in agricultural systems models. One overall critique from within economics as well as from other disciplines, is that the assumption of fully rational human decision making is often an unrealistic one. Alternatives to this assumption have also been employed in
agricultural systems models. For example, Dobbie and Balbi (2017) employ much simpler decision-making ‘heuristics’ or rules of thumb when modeling the human decision making done to allocate resources for their agent-based model of Malawian smallholders. These ‘heuristics’ may not generate optimal outcomes for the households, with respect to the economic concepts of utility or profits. However, this may
appropriate, because these outcomes may be closer to the outcomes achieved through actual decision making practices employed by individuals, particularly in light of limited information or cognitive limitations or bias in interpreting the information that is available.
Review of household model analyses of food security outcomes
We conducted a Scopus search of the search terms “Household Food Security Model” to identify the extent of existing research on food security modeling at the household level.
The initial Scopus search returned 997 references that model food security at the household level in a wide variety of ways. Across this initial set of works, we found three main categories of research on food security: research at the household level, in high income settings, without agriculture; low- and middle-income settings without agriculture and low- and middle-income settings with explicit reference to agriculture.
Although the first two categories are not of primary interest, these papers often discuss complex relationships between food security and other health and welfare outcomes of interest (like maternal and child health, HIV status and food security, food security in low-income urban areas etc.). Food security is either an outcome to be explained by a host of other factors (wages, demographics, poverty, living conditions or locations, for example), or as an explanatory factor for other outcomes, primarily health related (such as maternal and child nutrition, obesity). Due to the fact that these papers omit the supply side considerations of food production via the agriculture sector, they are considered outside of the scope of our review.
For our specific objectives, we focused on the third category of household food security analysis within an agricultural setting in a low- or middle-income region. Of the original 997 search results, 84 papers (detailed listing in Appendix Table A2) explicitly discussed both agriculture and food security. Despite the fact that these works explicitly
mentioned both food security and agriculture, not all works examined the linkages between agriculture and food security to the same degree. The overwhelming majority of papers utilized statistical methods with cross-sectional data to assess various causal relationships between food security and an agricultural variable of interest.
Furthermore, definitions of food security itself varied across these works, ranging from equating yields to food security directly, to utilizing one of the specific food security metrics we have identified as potential candidates for linking into agricultural systems research (like the Household Dietary Diversity Score, for example). Within this category, four broad categories of research were identified:
Papers that are motivated by issues of food security, but food security itself is not modeled. Food security is invoked in the motivation for the paper or in the abstract, but food security is implicitly equated to yields or increased productivity. Examples of this approach include analyses of vaccination rates for livestock (DeBruyn et al., 2017), adoption rates for drought tolerant maize varieties (Ali et al., 2017), women’s
empowerment programs (Burroway, 2016) and agricultural productivity (Haselow et al., 2016). No specific, validated food security metrics are used in these works.
One or more metrics representing a component of food security are analyzed as a function of a limited number of agricultural system level variables. Typically, the analysis in these papers makes use of an agricultural household survey (like an LSMS survey, for example) that has both a production and a consumption module, and possibly a distinct food security module, like HFIAS, included in the household survey.
This literature most often assesses statistical relationships between different
agricultural household production variables and food security status that is assessed with a specific food security indicator. Examples include the relationship between farm production diversity and household dietary diversity (Islam et al., 2018); farm size (area) and food security and food self-sufficiency (Waithaka et al., 2006) off-farm income prevalence and food expenditures (Zereyesus et al., 2017) coffee certification
